|
|
| |||||||||||||||||||||||||||||||||||||||||||||||
Background There is emerging concern that the methods used to measure the quality of care unfairly penalize providers caring for patients with multiple chronic conditions. We therefore sought to study the relationship between the quality of care and the number of medical conditions a patient has.
Methods We assessed measurements of the quality of medical care received in three cohorts of community-dwelling adult patients in the Community Quality Index study, the Assessing Care of Vulnerable Elders study, and the Veterans Health Administration project (7680 patients in total). We analyzed the relationship between the quality of care that patients received, defined as the percentage of quality indicators satisfied among those for which patients were eligible, and the number of chronic medical conditions each patient had. We further explored the roles of characteristics of patients, use of health care (number of office visits and hospitalizations), and care provided by specialists as explanations for the observed relationship.
Results The quality of care increased as the number of medical conditions increased. Each additional condition was associated with an increase in the quality score of 2.2% (95% confidence interval [CI], 1.7 to 2.7) in the Community Quality Index cohort, of 1.7% (95% CI, 1.1 to 2.4) in the Assessing Care of Vulnerable Elders cohort, and of 1.7% (95% CI, 0.7 to 2.8) in the Veterans Health Administration cohort. The relationship between the quality of care and the number of conditions was little affected by adjustment for the difficulty of delivering the care recommended in a quality indicator and for the fact that, because of multiple conditions requiring the same care, a patient could be eligible to receive the same care process more than once. Adjustment for characteristics of patients, use of health care, and care provided by specialists diminished the relationship, but it remained positive.
Conclusions The quality of care, measured according to whether patients were offered recommended services, increases as a patient's number of chronic conditions increases.
Quality indicators are typically developed for patients with one condition. Yet many patients have multiple conditions of varying severity. Sixty-five percent of Medicare beneficiaries have more than one condition, and almost 32% have four or more.11,12 Furthermore, physicians and systems provide care for patients with conditions that vary in complexity.
As interest in public reporting and pay-for-performance programs grows, many worry that providers who treat patients with more complex conditions will be unfairly penalized, with unintended consequences.13,14,15 Some prevailing concerns are that existing sets of performance measures do not adequately account for the number and severity of conditions16,17 or for the presence of specific conditions, such as major depression, that may affect the care of other conditions.18,19 Concerns about physicians having insufficient time to address the many needs of patients with multiple coexisting conditions have been bolstered by the results of simulation studies that show the mismatch between the time available to perform the vast number of recommended care processes and the time and personnel resources available to a typical primary care provider.20,21 Additional concerns are that the treatment of vulnerable subgroups such as homeless patients or patients with uncommon conditions, whose clinical needs and priorities may not be adequately represented in sets of performance measurements,22,23 will suffer if pay-for-performance programs create incentives that are misaligned with the needs of patients with complex conditions.
However, there is little empirical information about the influence of the number of chronic conditions on the quality of care. We therefore evaluated the variation in quality of care as the number of chronic conditions increases. We used clinically detailed data sets from three separate studies of various populations that measured disease-specific and aggregate quality of care for more than 7500 patients, 956 of whom had three or more chronic conditions.
Methods
Patient Populations
We analyzed data for patients in the Community Quality Index (CQI) study,4 the Assessing Care of Vulnerable Elders (ACOVE) study,24,25 and the Veterans Health Administration (VHA) quality-of-care project.26 A total of 6712 adults living in 12 metropolitan areas across the United States were enrolled in the CQI study between October 1998 and August 2000. They were asked to complete a telephone interview regarding their health care experiences in the 2 years before the interview and to provide written informed consent to obtain copies of their medical records from all providers. All patients provided such consent.
ACOVE recruited 372 vulnerable older patients, defined as community-dwelling persons aged 65 years or older who were at increased risk for death or functional decline, through two large managed-care organizations. Randomly selected patients were contacted by telephone between October 1999 and January 2000, and those who were determined to be at increased risk according to the Vulnerable Elders Survey–13 scale were included in that study.27 All patients provided written informed consent.
The VHA project involved 596 male veterans, aged 35 years or older, who were receiving care at 26 clinical sites in 12 systems of two Veterans Integrated Service Networks. These veterans received care from a participating VHA facility at least once in each of the 2 years between October 1997 and September 1999. The institutional review boards of the VHA waived the requirement for written informed consent.
The study protocols of the three projects were approved by the RAND institutional review board. The VHA project was approved by the institutional review boards of individual VHA facilities.
Measurement of the Quality of Care
In all three projects, the quality of care was measured with the use of quality indicators related to care processes.4,24,26 Each quality indicator specified which patients were eligible and which care processes the eligible patients should receive. These quality indicators were developed on the basis of systematic reviews of the medical literature and formal judgments by an expert panel. The CQI study involved 439 quality indicators covering 30 clinical conditions and preventive care.28 The VHA project used the 348 quality indicators from the CQI study that were relevant to the care of male veterans, covering 26 clinical areas.26 The 236 quality indicators in the ACOVE study specifically targeted vulnerable older patients, covering 22 medical and geriatric conditions and crosscutting clinical areas.25 Examples of the quality indicators are presented in Table 1. Information on medical care provided to the study patients was derived from medical records and supplemented by interviews with patients in the CQI and ACOVE studies about selected quality indicators for which the patient was judged to provide information that was more reliable than the medical records. Medical records were abstracted by nurses with previous experience in medical-record abstraction who also underwent rigorous training. Abstraction of data used to determine the satisfaction of quality indicators was uniform, as indicated by kappa statistics of 0.80 or higher.
|
Medical Conditions
The primary predictor variable in this study was the total number of identified conditions affecting each patient. We selected chronic conditions that require continuous management, and we studied both common and unique sets of conditions. The conditions common to patients across the three studies were depression, diabetes mellitus, heart failure, stroke, hypertension, coronary artery disease, osteoarthritis, chronic obstructive pulmonary disease, and atrial fibrillation. Conditions unique to patients in the ACOVE study were dementia, pressure ulcer, osteoporosis, urinary incontinence, and renal insufficiency. Conditions unique to patients in the CQI and VHA projects were benign prostatic hypertrophy, dyspepsia, asthma, colorectal cancer, and prostate cancer. Breast cancer was present only in the CQI cohort.
Statistical Analysis
Main Analysis
We first plotted the relationship between the number of conditions a patient had and the overall quality of care the patient received. After confirming the linear trend, we ran linear regression models at the patient level, using quality of care as the outcome variable and the number of conditions as the primary predictor variable. The coefficient of the number of conditions was used to estimate the increase in the quality of care associated with the addition of a coexisting condition.
There were at least two types of potential artifacts in the main statistical analysis. The first arose from the fact that patients were eligible for various sets of quality indicators (according to their medical conditions) and that the difficulty of delivering the care recommended in an indicator could vary.25 To adjust for this, we calculated an expected quality score, on the basis of population scores, for the quality indicators for which the patient was eligible.29 We subtracted the expected score from the patient's actual score to calculate a quality score adjusted for difficulty. We then examined the relationship between the adjusted quality score and the number of conditions the patient had.
A second potential artifact was that the same care processes satisfy quality indicators that target different conditions. For example, daily antiplatelet therapy is recommended both for patients with diabetes and for patients with coronary heart disease. Therefore, patients with both conditions are eligible for aspirin therapy twice. To control for this artifact, we counted multiple eligibilities for the same indicator only once in the score calculation.
Exploratory Analysis of Mechanisms of the Quality–Care Relationship
To explore potential mechanisms of the observed relationship between the number of conditions and the quality score, we tested the effect of three potentially contributing factors: demographic characteristics of patients, use of health care, and care provided by specialists. After adjusting for demographic characteristics and use of health care in the model, we examined the change in the slope of the regression curve for the quality of care and number of conditions. Quality scores for the ACOVE and CQI data were adjusted for age, sex, education level, annual income, self-reported health, and geographic characteristics (i.e., the two managed-care organizations in the ACOVE study and the 12 metropolitan areas in the CQI study). The CQI data were further adjusted for race and ethnic group and for type of health insurance. We did not adjust the ACOVE data for these variables, because the patients were predominantly white, and all were enrolled in a managed-care plan for seniors. All patients from the VHA project were male, and their other demographic data were limited; therefore, we adjusted only for age and geographic region (i.e., Veterans Integrated Service Network). The adjustment for the use of health care addressed the number of office visits in each of the three data sets and the number of hospitalizations in the CQI and ACOVE data sets.
Using the ACOVE data, we examined the effect of the involvement of specialists and the number of providers. Data on the specialty of providers were not collected in the CQI and VHA projects. We performed the first analysis by consecutively adding two variables to the linear-regression model of quality score and number of conditions: first, a binary variable indicating whether or not the patient received any specialist care, and second, the interaction term for the binary variable and the number of conditions. The interaction term was used to examine the difference in the relationship of quality score to the number of conditions between patients who were treated only by generalists and patients who received any care from specialists. We conducted the second analysis by adding data on the number of providers to the linear regression model adjusted for demographic characteristics.
All analyses were conducted with the use of Stata software, version 9.2. P values less than 0.05 were considered to indicate statistical significance.
Results
Patient Populations
Table 2 describes the characteristics of the three populations of patients. The numbers of conditions ranged from zero to eight among patients in the ACOVE study and from zero to seven among patients in the CQI and VHA projects. Patients were eligible for a mean of 20, 16, and 19 quality indicators in the ACOVE, CQI, and VHA data sets, respectively. The mean quality scores were 55% in both the ACOVE and CQI cohorts and 67% in the VHA cohort.
|
Figure 1 shows the relationship between the number of conditions and the quality of care received in each of the three cohorts. In general, as the number of conditions increased from zero to five, the quality-of-care scores also increased; small numbers of patients with six or seven conditions resulted in imprecision of those point estimates. The unadjusted linear regression analysis showed that each additional condition was associated with a 2.2% increase in the quality score (95% confidence interval [CI], 1.7 to 2.7) for the CQI cohort, 1.7% (95% CI, 1.1 to 2.4) for the ACOVE cohort, and 1.7% (95% CI, 0.7 to 2.8) for the VHA cohort. Figure 2 shows the effect of adjustment for two potential artifacts of measurement on the slope of the relationship between the quality of care and number of conditions. Since the effect was minimal, later analyses did not involve such adjustment.
|
|
Figure 3 presents the effect of adjustment for characteristics of patients and use of health care. For the ACOVE and CQI cohorts, the relationship between the quality score and the number of conditions remained moderately positive after adjustment; in the VHA cohort, the adjusted slope was positive but not significantly different from a slope of zero (P=0.17). The adjustment for care provided by specialists and number of providers was possible only in the ACOVE data set. Among the 372 ACOVE patients who were eligible for 10,711 care processes, 233 saw at least one specialist for a condition for which there was a quality indicator. Addition of the binary variable indicating whether or not the patient received any care from a specialist did not change the relationship between the number of conditions and the quality score. The analysis including the interaction term between receipt of care from a specialist and number of conditions showed that the relationship between the number of conditions and the quality score was significantly smaller (but still positive) among patients who did not receive care from specialists than among those who did, although this difference was significant only in the unadjusted analysis (the change in the slope of the regression curve, 0.6 vs. 2.2 percentage points for the unadjusted analysis, P=0.02; 0.73 vs. 2.0 percentage points for the analysis adjusted for demographic characteristics, P=0.07; and 0.49 vs. 1.8 percentage points for the analysis adjusted for demographic characteristics and use of health care, P=0.08). The addition of the number of providers to the model adjusted for demographic characteristics minimally reduced the magnitude of the increase in the slope of the quality–condition curve (from 1.60 percentage points for each additional condition to 1.54 percentage points). Each additional provider involved was associated with an increase of 0.3 percentage point (P=0.39).
|
Contrary to our expectations, we found that the percentage of care recommended in quality indicators that was delivered to each patient increased, rather than decreased, as a patient's number of chronic conditions increased. The fact that we found essentially the same relationship in three different data sets, using two different sets of quality indicators, increases the likelihood that this effect is real, rather than an artifact of any one particular study. This finding does not provide support for the argument that incentive programs based on quality indicators of care processes will necessarily penalize providers who provide care to patients with multiple conditions.
Unlike our study, previous studies that have examined the effect of coexisting conditions on the provision of necessary care processes have generally focused on single areas of care, such as preventive care, with mixed results. Redelmeier and colleagues30 reported finding a negative effect of the presence of diabetes on receipt of estrogen-replacement therapy, of pulmonary emphysema on the necessary use of lipid-lowering medications, and of psychotic syndromes (as identified by the use of haloperidol) on treatment for arthritis. Desai and colleagues31 found little difference in the care received for diabetes among veterans with mental disorders and those without. Heflin and collegues32 found an increased likelihood of undergoing a clinical breast examination, Papanicolaou smears, and fecal occult-blood testing among older patients with hypertension, whereas the likelihood of mammography was reduced among patients with hip fracture. They found no evidence that patients with other conditions, such as stroke and diabetes, had a reduced likelihood of undergoing four cancer-screening processes.
Although we had limited data on the mechanisms underlying the positive relationship between the quality score and the number of conditions, several of our analyses were informative. First, the increased use of health care by patients with more conditions explained some of the positive quality–condition relationship. In all three data sets, the slope of the curve was less after we adjusted for the use of health care than before adjustment. This effect was particularly prominent in the VHA data set, for which the slope was essentially flat after adjustment. This finding does provide support for the hypothesis that patients with more conditions receive higher-quality care than patients with fewer conditions, because they have more opportunities to receive care. Second, the data from the ACOVE study suggest that the involvement of specialists plays a modest role in the positive relationship between the number of conditions and the quality of care. The quality score for each additional condition increased more for patients who had seen a relevant specialist than for those who had not. This result is consistent with reports in the literature that specialists provide higher-quality care in their areas of specialty than do generalists.33,34,35 However, for patients who received only generalist care, the relationship between the quality score and the number of conditions remained positive. These data suggest that generalists can provide equivalent care to patients with complex conditions and to those with less complex conditions. A potential mechanism that we could not assess in our study, however, is the possibility that seeing a specialist is a marker for patients who advocate more effectively for the care they need.
Our study had several limitations. First, we used a simple count of conditions as the primary predictor variable. This variable is a crude measure of complexity, since clinicians do not view all coexisting conditions as equivalent. For example, they view depression as a more challenging coexisting condition than hypertension or osteoarthritis. Furthermore, our data do not take into account the severity of illness for any particular condition, such as severe diabetes as compared with mild diabetes. The data sets were, in general, too small to permit the examination of specific positive and negative effects on the quality of care processes for various combinations of conditions. Because we considered only conditions that had corresponding quality indicators, we underestimated the total disease burden. Our sets of quality indicators did not assess psychosocial issues, which may be an additional contributor to the difficulty in providing quality care. However, the sets were extensive, covering the most prevalent conditions.
Second, the quality score appeared to decrease for patients in the CQI and VHA cohorts who had seven conditions (five patients [<0.1%] and three patients [0.5%], respectively), although the small samples prevent us from making any conclusions for these patients. Future studies should focus on the quality of care provided to patients with a highly complex disease burden, although such patients are likely to account for a very small fraction of any provider's patients. Third, since our performance measures focused on technical quality of care (whether or not patients were offered recommended services), we cannot draw conclusions about the relationship between the number of conditions and the experience of the patient, the quality of communication between the patient and physician, or the outcomes of care. Finally, the CQI, ACOVE, and VHA projects measured quality with the use of a comprehensive set of care processes that were carefully constructed to account for a patient's detailed clinical status. Our results may not be generalizable to quality indicators based on outcomes or quality indicators that are less clinically detailed or that assess a smaller percentage of the care received.
In summary, contrary to expectations, we found that patients with more conditions that have corresponding quality indicators consistently received better technical quality of care for these conditions than did patients with fewer conditions. This finding suggests that comprehensive, clinically detailed sets of care processes received can be used to assess the quality of care without creating a disincentive for providers to avoid patients with the most prevalent chronic conditions.
Supported by the RAND Corporation and the National Primary Care Research and Development Centre, England.
No potential conflict of interest relevant to this article was reported.
Source Information
From Kyoto University, Kyoto, Japan (T.H.); University of California at Los Angeles (N.S.W.) and the Veterans Affairs Greater Los Angeles Healthcare System (C.F., S.M.A., P.G.S.) — both in Los Angeles; RAND Health, Santa Monica, CA (N.S.W., J.L.A., C.F., E.A.M., S.M.A., P.G.S.); National Primary Care Research and Development Centre, University of Manchester, Manchester, United Kingdom (M.R., D.R.); and the Veterans Affairs Ann Arbor Health Care System and the University of Michigan Department of Internal Medicine — both in Ann Arbor (E.A.K.).
Address reprint requests to Dr. Shekelle at the RAND Corporation, 1776 Main St., P.O. Box 2138, Santa Monica, CA 90407-2138, or at shekelle{at}rand.org.
References
| |||||||||||||||||||||||||||||||||||||||||||||||
Related Letters:
Number of Medical Conditions and Quality of Care
Steinman M. A., Higashi T., Wenger N. S., Shekelle P. G.
Extract |
Full Text |
PDF
N Engl J Med 2007;
357:1350-1351, Sep 27, 2007.
Correspondence
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2009 Massachusetts Medical Society. All rights reserved. |
Previous
